Dynamics Dynamics

Abstract. During the Arctic Research of the Composition of the Troposphere from Aircraft and Satellites (ARCTAS) campaign, two NASA research aircraft, a DC-8 and a P-3B, were outfitted with extensive trace gas (the DC-8) and aerosol (both aircraft) instrumentation. Each aircraft spent about a half hour sampling air around the oil sands mining and upgrading facilities near Ft. McMurray, Alberta, Canada. The DC-8 circled the area, while the P-3B flew directly over the upgrading plants, sampling close to the exhaust stacks, then headed downwind to monitor the aerosol as it aged. At short range, the plume from the oil sands is a complex mosaic of freshly nucleated ultrafine particles from a SO2- and NO2-rich plume, soot and possibly fly ash from industrial processes, and dust from dirt roads and mining operations. Shortly downwind, organic aerosol appears in quantities that rival SO4, either as volatile organic vapors condense or as they react with the H2SO4. The DC-8 pattern allowed us to integrate total flux from the oil sands facilities within about a factor of 2 uncertainty that spanned values consistent with 2008 estimates from reported SO2 and NO2 emissions, though there is no reason to expect one flyby to represent average conditions. In contrast, CO fluxes exceeded reported regional emissions, due either to variability in production or sources missing from the emissions inventory. The conversion rate of SO2 to aerosol SO4 of ~6% per hour is consistent with earlier reports, though OH concentrations are insufficient to accomplish this. Other oxidation pathways must be active. Altogether, organic aerosol and black carbon emissions from the oil sands operations are small compared with annual forest fire emissions in Canada. The oil sands do contribute significant sulfate and exceed fire production of SO2 by an order of magnitude.

issues such as soil acidification, trace metals in foods harvested by Aboriginal communities, and vegetation stress. Data from many of these sites can be accessed directly through the WBEA website. One of the unusual features of the oil sands facilities is their location in an area lacking large populations and other industrial facilities. This makes the emissions more 5 obvious against a relatively clean background. Hence, the oil sands emissions can be discerned in global satellite maps of NO 2 and SO 2 (McLinden et al., 2012) even though total emissions are smaller than many urban regions. This input of pollutants into a nearly pristine area has been of concern since development began in the area. Studies of aerosol formation (Cheng et al., 1987), deposition (Barrie, 1980;Proemse 10 et al., 2012b) and composition (Proemse et al., 2012a) have been ongoing as have air quality modeling efforts (Davies, 2012).
Despite this emphasis on measuring oil sands emissions, there had been no recent evaluations of them using an extensive airborne instrumentation suite until the summer of 2008, when the NASA DC-8 and P-3B research aircraft were deployed at the Cana-15 dian Forces Base Cold Lake in Alberta, Canada. The main focus of the project was to study smoke plumes from the forest fires that occur every year in Northern Canada (Jacob et al., 2010), but the two planes detoured one day to measure aerosol and trace gas emissions from the facilities near Ft. McMurray.
The gas phase data revealed two different types of plumes (Simpson et al., 2010

Experimental
The flights around the Ft. McMurray oil sands upgrading facilities took place on 10 July 2008. The DC-8 and P-3B took the opportunity to do a joint flight over the oil sands to perform an instrument intercomparison and a brief evaluation of the aerosol and gas emissions from the facilities there (Fig. 1). The DC-8 flew a loop around the 5 facilities at about 350 m a.g.l., passing through the entire plume and providing upwind samples for contrast. The P-3B flew a figure 8, passing through visible plumes from the tallest stacks in the facilities (Fig. 2). The Syncrude plume was sampled twice at about 240 m a.g.l. and once at 625 m while the Suncor plume was sample once, at about 250 m. After the third Syncrude plume penetration, the P-3B turned downwind 10 and followed the evolving plume.
In addition to the 10 July flights, there were 3 plume penetrations on other ARCTAS flights (Fig. 3). The P3-B descended into the mixed layer during transits from fire plume studies on 28 and 29 June, while the DC-8 flew a low pass early on 28 June looking for a fire plume from the previous day and coincidentally met the oil sands plume.

Instrumentation
The two aircraft had complementary payloads. The DC-8 was outfitted with an extensive suite of gas-phase and aerosol instruments. The P-3B had similar aerosol measurements and several radiation propagation measurements, but no gas phase capability beyond CO and O 3 . Both aircraft used the aerosol inlet characterized by Mc-20 Naughton et al. (2007), who found that it efficiently conveyed dust particles of up to a few microns in diameter. Additional information and data from ARCTAS can be found at [http://www.espo.nasa.gov/arctas/airborne_inst.php]. Rather than describing the entire set of measurements, a short description of the most relevant ones is included here. Introduction with combustion, as flames produce nonvolatile black carbon and organic material, though dust, seasalt, and volatile particles too large to evaporate entirely can also contribute (Clarke, 1991). Differences between UCN and CNcold show particles in the 3 to 10 nm range indicative of recent new particle formation (nucleation). UCN and CNcold suffered from saturation of the counters-concentrations so high that multiple particles 10 were in the sensing volume simultaneously and were not counted properly. The counters used for CNcold and CNhot do not compensate for coincident particles, which becomes a problem at > 10 × 10 3 cm −3 . A correction algorithm was applied (Baron and Willeke, 2001), but the algorithm fails above about 35 × 10 3 cm −3 , when particles are in the sampling volume ≥ 50 % of the time. The UCN counter automatically compensates 15 for coincident particles, but saturates at 100 × 10 3 cm −3 . Aerosol size distribution measurements differed modestly between the platforms. Both used Differential Mobility Analyzers (DMAs, TSI long DMA model 3081) for particles from ∼ 10 to 400 nm, Optical Particle Counters (OPCs) for rapid measurement of the the accumulation mode (0.2 to 1 µm), and Aerodynamic Particle Sizers (APS, TSI 20 model 3321) to measure aerodynamic particle diameters between ∼ 0.8 µm < D ae < 10 µm. The DC-8 used a DMT UHSAS OPC, which resolves smaller particles (0.055 to 1 µm) at a higher rate (1 Hz) than the modified LAS-X OPC on the P3-B (0.2 to 3 µm, 0.33 Hz), though the latter had a set of heated inlets that could explore volatility at 150, 300, and 400 • C (residence times > 1 ). The P-3B also had a TSI radial DMA with Introduction  (DeCarlo et al., 2006). The P-3B AMS ran on an approximately 40 % duty cycle, blanking 2 s, sampling for 2,s, and taking about a second to prepare for the next cycle. The DC-8 ran in so-called "fast mode" (Kimmel et al., 2011)  (OA) mass concentrations extracted using the publicly available Squirrel software (Allan et al., 2003(Allan et al., , 2004DeCarlo et al., 2006) and the collection efficiencies described in Middlebrook et al. (2012). It should be noted that he AMS has difficulty separating organosulphates from inorganic sulphate (Farmer et al., 2010). Black Carbon (BC) was measured with DMT Single Particle Soot Photometers (SP2s, Schwarz et al., 2006;10 Stephens et al., 2003). Unfortunately, the DC-8 SP2 was not operational on 10 July. The DC-8 had additional aerosol composition measurements, but at rates too slow to capture the oil sands plume. The aerosol optics packages on both aircraft included TSI model 3563 nephelometers (Anderson et al., 1996) measuring light scattering B sp at 450, 550, and 700 nm and 15 Radiance Research Particle Soot Absorption Photometers (PSAPs) obtaining light absorption B ap at 470, 530, and 660 nm. Nephelometer data were corrected for truncation errors using the procedures from Anderson and Ogren (1998). PSAP corrections were performed according to Virkkula et al. (2005) and Virkkula (2010).
As mentioned above, the DC-8 had an extensive set of trace gas measurements. Of 20 particular interest here are CO, SO 2 , NO 2 , NO x , and a suite of hydrocarbons and halocarbons collected in stainless steel chambers and measured with gas chromatography at UC Irvine. More extensive gas phase data are reported in Simpson et al. (2010). The DC-8 was equipped with a DIfferential Absorption Lidar (DIAL, Browell et al., 1998;Dupont et al., 2012) which provides a 2-D curtain of backscatter data above and Introduction for an atmosphere lacking aerosol L m .
where β p and β m are backscattering due to particles and molecules (air) respectively, and ε p is extinction along the beam due to particles. For the qualitative analysis here, this approximation is sufficient, but note that R ba is an underestimate of the actual 5 backscatter ratio when ε p is significant, which often occurs when the DIAL penetrates thin clouds.

Results and discussion
The 10 July 2008 flights provide the bulk of our data. It was a clear day with broken low clouds and some higher clouds in the distance (Fig. 2). Surface winds below 1 km 10 were from the SSW at 3 to 7 m s −1 (near the climatological mean) but increased with altitude to 15 m s −1 at 1 km and about 20 m s −1 at 3.5 km. Above 1 km winds transitioned to about 215 ± 15 • . Convective activity and clouds were present in the area but no precipitation was evident. Some clouds interacted with emission plumes apparently mixing them to higher altitudes.

Near-field characterization
The closest plume penetrations were too rapid for the PSAPs and size distributions, but the rapid-response measurements showed the plumes clearly (Fig. 4). SO = 4 and CO were found almost exclusively in the visible plumes from the tallest stacks. In contrast BC and OA appear to be produced in other locations as well, probably including Introduction the plumes from the upgrading facilities are spatially heterogeneous, with a variety of plume compositions. The behavior of organic aerosol after the last penetration is striking. Only a small amount was emitted directly -there were only small fluctuations during the close passes -but as soon as the P-3B turned downwind, OA concentrations jumped to 5 almost equal SO = 4 ( Fig. 4 after 20:15 UTC) and the roughly 1 : 1 ratio between OA and SO = 4 persisted for the rest of the leg. The origin of the organic matter is not clear; the high correlation with SO = 4 indicate that the tall stacks are also the OA source, but it is conceivable that it was mixing up into the SO = 4 plume from lower altitude sources. Perhaps the extensive vapor phase measurements of the DC-8 could distinguish between 10 those possibilities, but the P-3B payload lacked that capability. It is also not immediately apparent whether the organic aerosol was simply condensation of semivolatile vapors as the plume cooled with altitude (and is thus primary OA) or whether photochemical reactions could have produced secondary organic material in the short period since emission. Elemental analysis of the AMS mass spectra revealed median O : C = 0.26 15 and H : C = 1.5, characteristic of fresh mixed primary and secondary OA (Aiken et al., 2008), which is also consistent with the mass spectral pattern observed by Ng et al. (2011).
It is likely that reactions with H 2 SO 4 are responsible for much of the OA. The high acidity of the plume could catalyze polymerization of aldehydes (Jang et al., 2002), 20 react with alkenes such as isoprene (Surratt et al., 2007) or react with alcohols such as 2-methyl-3-buten-2-ol (Zhang et al., 2012). While much of the organic vapor in the oil sands plume was aliphatic and thus unreactive, alcohols and alkenes such as isoprene, α-pinene, and β-pinene were enhanced (Simpson et al., 2010). Biogenic vapors from the surrounding forest may also have contributed. 25 As part of the downwind leg, the P-3B explored the vertical structure of the plume by ascending through the mixed layer and the cloud layer, then profiling back down to the mixed layer. Cloud base was about 2.2 km. During the ascent, clouds were avoided, but pulses of aerosol were clearly associated with close approaches. Similar pulses were present in and between clouds during the descent (blue regions in Fig. 4). It is not clear how much of the sulphate in and around clouds was due to the well-known aqueousphase reaction of SO 2 with H 2 O 2 (Penkett et al., 1979). In this case the reaction is limited by the roughly 500 pptv concentration of H 2 O 2 seen by the DC-8, which could produce as much as 2 µg m −3 SO  Littlejohn et al., 1993;Sarwar et al., 2013), though this is a pH-dependent reaction that may be limited by available NH 3 (which was not measured). Figure 5 shows a similar dataset from the DC-8. At the 10 km distance of this sampling, the plume is more mixed, but some of the same features are visible. There is 10 a double hump structure to many of the species. Given the winds and flight direction, the first peak would be from Suncor and the second from Syncrude. As with the P-3B, SO = 4 was found in a narrower region, indicating a couple of discrete sources. SO 2 behaved similarly, while other species were more widely spread out. NH + 4 was also higher in the plume, but not sufficiently to neutralize the SO background air to 0.5 in the heart of the plume, indicating that NH 3 was limited and the plume was acidic. Before the main plumes, from 18:25 to 18:28 UTC, scattering, absorption, OA, CO, CO 2 , and CNhot were all moderately enhanced, due either to mining or the city of Ft. McMurray. In contrast, NO 2 was quite low during that period.
Size distributions in the plume seen by the DC-8 are complex (Fig. 6). The APS 20 (top panel) shows modes at 3 µm at 18:30 and 18:32 UTC. That is typical for dust made mechanically and is likely to be road dust, from mining, or from crushing or mixing gravel. The 1 µm mode around 18:31:40 UTC is more unusual -too small for mechanical generation and much larger than the typical accumulation mode of 0.1 to 0.3 µm from gas to particle conversion. The latter is also seen in the UHSAS data at ACPD 13,2013 Particulate emissions from Athabasca oil sands S. G. Howell et al.  (Barrie, 1980). Total and submicrometer light scattering is also plotted in the top panel. It is no surprise that coarse mode scattering is high during the dust events. However, maximum scattering occurred with the suspected fly ash, since 1 µm particles are near the peak of mass scattering efficiency. Figure 6 also reveals that near 18:30:30 UTC the 0.1 to 0.2 µm accumulation mode 5 caused 80 % of the scattering. This is coincident with the SO = 4 and OA peaks from the AMS (Fig. 5). These sizes are very effective as CCN in BL clouds. CNhot, which typically consists mainly of BC and refractory organics (Clarke et al., 2007) is enhanced during the same period but also shows narrow peaks at other times (e.g. near 18:30 UTC). This is similar to those seen in the P3B data for stack emissions with high 10 BC ( Fig. 4). Consequently, the size distributions indicate different sources or processes embedded in the broad oil-sands plume contribute differently to both coarse and fine aerosol.

Vertical structure
Although aircraft have the ability to survey in 3 dimensions, in-situ measurements are 15 limited to a 1-D line of observations along the flight track. A lidar curtain is an effective way to add a dimension, and thus context, to aerosol data. However, because aerosol size depends upon water uptake in response to relative humidity that generally varies with altitude (Shinozuka et al., 2011), the backscatter cannot directly quantify emitted dry aerosol mass. The DC-8 lidar had difficulty penetrating clouds in some spots, but 20 did provide an overview of the plume (Fig. 7). Figure 7b relates the lidar curtain and location of the plume to winds and geography. The wind vectors don't line up precisely with the 18:03 UTC start of the plume or maximum at 18:06 UTC as seen by the lidar, but small changes in wind direction are likely in the few hours it took the plume to reach the lidar track. The backscatter 25 peak between 18:10 and 18:11 UTC is probably due to mining operations (see Fig. 1 downwind of the city of Ft McMurray. Aerosol in the main plume appears to be mixed fairly uniformly to about 1450 m, but is drawn higher by convection around 18:05 UTC, presumably in a manner similar to that seen by the P-3B among clouds in Fig. 4.

Fluxes
One of the advantages of sampling from a mobile platform is that one can cross the 5 entire plume and integrate a total flux if the vertical structure is known. The flux in the direction of the wind through the vertical projection of the flight track between times t 0 and t 1 is where z m is the top of the boundary layer, S and S 0 are the plume and background 10 concentrations, v a and φ a are aircraft speed and heading in Earth coordinates, and v w and φ w are the wind speed and direction. If we assume winds and concentration are constant with altitude throughout the mixed layer and there is no vertical flux above the mixed layer then the discrete version of Eq.
(2) becomes The flux from a source equals that through the flight path curtain if there are no significant sources or sinks along the way. The DC-8 loop around the two major upgraders is well suited to such a calculation. The upwind part of the loop provides the background concentrations (blue areas of no vertical profiles in the plume, so we have no direct measurements of mixed layer depth there. Figure 8 shows vertical profile data from the lidar and in-situ instruments. The lidar shows a fairly uniform mixed layer topping out at 1000 to 1200 m. In contrast, the potential temperature suggests a very stable boundary layer with very little vertical mixing. Water vapor and CN concentrations are not uniform, but have minima near 5 1200 m. Since the ground level is about 300 m, we use a mixed layer depth of 900 m, but it could be as small as 700 or as large as 1500 m (the P-3B profile about 100 min later showed CO capped at about 1800 m). Table 1 shows the results of the flux calculations for both mixed layer depths. We note that the AMS SO = 4 flux is roughly 5 % of the SO 2 flux roughly an hour downwind of the 10 upgraders, consistent with Cheng et al. (1987), who found that SO 2 from the upgraders reacted at about 6 % h −1 . This conversion rate cannot be explained by reaction with OH alone. The OH concentration measured on the DC-8 was slightly elevated, averaging about 0.11 pptv across the plume except at the SO 2 peak, when it dropped to 0.05 pptv. Background concentrations averaged 0.07 pptv. At 0.1 pptv OH, 286 K, and 92 500 Pa 15 the reaction rate coefficient is 1.06 × 10 −12 cm molecule −1 s −1 (Blitz et al., 2003), consuming 310 pptv SO 2 h −1 , or about 1 % h −1 . If SO 2 from the stacks is the only source of SO = 4 , reaction with OH is less than 20 % of the total SO 2 →SO = 4 conversion. Since the SP2 was not functioning on the DC-8, we used absorption data from the sub-µm PSAP. Particulate mass absorption efficiency (MAE) is usually reported as be- 20 tween 5 and 20 m 2 g −1 BC, depending on the geometry of the soot and its coatings (Fuller et al., 1999). The P-3 had functioning PSAPs and SP2 that day, so MAE immediately adjacent to the upgraders could be determined, as shown in gray in Fig. 9. Water vapor fluctuations in the biggest plumes caused large positive and negative artifacts in the PSAP signals that had to be edited out, but BC was primarily found outside 25 those plumes so the average should be valid. The data are entirely consistent with the MAE of 7.5 ± 1.2 m 2 g −1 for uncoated soot found by Bond and Bergstrom (2006). However, rapid changes in particle composition as the plume ages are likely to affect MAE, so the P-3B flyby on 29 June is probably more representative of the 10 July DC-8 data, as both were about 10 km downwind. As seen in black in Fig. 9, the data, while scattered and a bit sparse, indicate MAE is roughly double that of the 10 July close pass. This amplification is as expected for particles with soot cores surrounded by nonabsorbing coatings (Bond et al., 2006) and is similar to laboratory studies of coated soot (Schnaiter et al., 2005). Table 2 enumerates most of the errors. "Instrument gaps" refers to how well the instrument can sample plume extremes. The AMS, for example, is blanking 25 % of the time so might miss peak concentrations. Convection out of the mixed layer is clearly visible in the lidar curtain (Fig. 7) between 18:04 and 18:06 UTC and by the P-3B when it is over 2.3 km (Fig. 4) but is sporadic and thus hard to quantify. Here we simply assume a maximum error caused by 25 % cloud cover with 1 m s −1 updrafts under those clouds exhausting mixed layer concentrations and returning zero concentration air between clouds. That would be sufficient to replace about half of the mixed layer in an hour. Dry deposition losses are not included because they are relatively small; even a deposition velocity of 1 cm s −1 would remove < 4 % of the column amount in the hour 15 between emission and the DC-8 sampling. NO 2 and SO 2 also have photochemical sinks, but ≤ 6 % h −1 for SO 2 (Cheng et al., 1987).
Environment Canada maintains a public website with historic emissions information. A search for SO 2 emissions in 2008 reveals that the Syncrude and Suncor plants at Fort McMurray together released about 1 × 10 7 kg in 2008, or 3100 g s −1 . Given our 20 estimated errors and the questionable assumption that SO 2 production on 10 July were representative of annual emissions, the combined SO 2 and SO = 4 fluxes of 4800 g s −1 is excellent agreement. NO y is even closer. The dramatic differences between reported and measured fluxes for CO are thus something of a surprise. CO 2 fluxes were also higher than expected, but by a factor of 2.5. It may be that CO and CO 2 production 25 are more variable than NO 2 and SO 2 or that major sources were not included in the release inventories. The largest uncertainties in the flux calculation are amenable to improvement with a more comprehensive flight plan. Multiple legs in the mixed layer and at higher alti- tudes would permit better quantification of mixed layer uniformity, height, and exchange with air above. Perhaps most valuable would be a LIDAR curtain directly over the fluxmeasurement legs, which could address all of those at once. Because they are unaffected by dilution, emission ratios can be determined with better precision than fluxes. Table 3 shows emission ratios of the aerosol species to 5 CO from this work and from fresh forest fire plumes during ARCTAS. At the 10 km DC-8 distance from the source, emission ratios were similar for OA, but BC is enhanced in the oil sands plume about 3 times more and SO = 4 :CO is higher by a factor of 50. However, the average CO enhancement in the oil sands plume averaged 10.3 ppbv and peaked at 98 ppbv while forest fire peaked an order of magnitude higher, so OA in 10 fire plumes often reached hundreds of µg m −3 (Cubison et al., 2011;Hecobian et al., 2011). Similarly, BC in the oil sands plume is small compared with even a single fire plume.
Comparisons between the oil sands plume and other aerosol sources suffer from some ambiguity, as results are strongly dependent on the time and space scales cho-15 sen. As an arbitrary example, Table 4 shows estimated annual emissions from forest fires throughout Canada from Amiro et al. (2009) in comparison to the oil sands fluxes from Table 1. On this scale the oil sands are an insignificant source of BC, OA, and CO (even though we found much higher CO than expected), but dwarf fire production of SO 2 . Two caveats should be noted here: BC and OA have no precise chemical defini-20 tion, so there are likely to be components of each that are primarily from the oil sands; and the fire season lasts only a couple of months, so aerosol from oil sands processing is certainly a more significant fraction of the total burden in other seasons.

Aerosol evolution
The P-3B and DC-8 flights near the sources on 7 July were the primary measurements 25 for the oil-sands emissions. However, we also encountered the plume downwind at other times as shown in Fig. 3. These "encounters of opportunity" provide data to examine plume development over time and some evolution in gas and aerosol properties.  Figure 10 shows P-3B data for the 29 June and 28 June flight segments, about 10 and 180 km downwind of the upgraders. Unfortunately, on 29 June, the CO went through a calibration cycle at exactly the peak of the plume. The SO = 4 peak that day was much smaller than on 10 July, indicating that the plane may have missed the main part of the plume or that SO 2 production was lower that day. As with the DC-8 data 5 discussed above and collected at roughly the same distance, the plume in Fig. 10a is complex, indicating a variety of sources. OA appears to have a much higher background, perhaps of biogenic origin as BC was low. A broad plume with dust and organic aerosol from 22:38 to 22:40 UTC may correspond to mining operations and the evaporative plume from Simpson et al. (2010). The enhanced SO = 4 and very large UCN 10 around 22:42 UTC indicate the upgrader plume. As with the 10 July data, most of the particles are volatile, so are not detected as CNhot. A review of the forward video reveals that the very narrow peak at 22:43:39 UTC occurred shortly after the plane flew over a road where trucks could be seen raising clouds of dust. Dust was responsible for much of the light scattering seen throughout the plume crossing; the very similar 15 patterns at all wavelengths can only occur with particles large compared to the wavelength. While SO = 4 was far smaller than that seen by the DC-8 a similar distance away, the dust seen by the APS was similar, though the 1 µm mode was absent, further evidence that the plane either missed part of the plume or that the industrial facilities were operating differently. 20 In contrast to the detail visible at 10 km, the plume at 180 km downwind (Fig. 10b) is broad and smooth. Wind speeds were about 10 m s −1 , so the plume was about 5 h old. The dust is no longer obvious, presumably diluted, as dry deposition time scales are likely to be about 12 h for 5 µm particles, which have a deposition velocity of roughly 2 cm s −1 in coniferous forests with that wind speed (Zhang et al., 2001). However, the 25 roughly 1 µm possible fly ash mode is evident. The disparity between the red and blue channels reveals that small particles now dominate the scattering. While overall particle concentrations are lower (note the log scales on the CN data), there are now very few volatile particles, and higher CNhot suggests that at least some of the secondary organic material in the aerosol has low volatility. Once again, SO = 4 and OA are well correlated. CO, which is stable at these time scales, rises about 6 ppbv.
Flux calculations like those in Sect. 3.3 can be be applied here, though errors are even more poorly quantified: the mixed layer top of 2900 m is solely from a descent prior to the plume crossing; the P-3B wind speeds were compromised by the radiometer on the nose of the aircraft; and the end time of the plume is rather ambiguous. Thus, these fluxes are just rough approximations. Table 1 shows the range of fluxes calculated for reasonable definitions of the end of the plume. Values for SO = 4 and OA do not change much in this calculation, while BC and CO are sensitive to the plume duration and are particularly uncertain.

10
While very approximate, the fluxes calculated in this flyby do confirm the high CO emissions from the DC-8 loop. The increased SO = 4 : CO is as expected, since sulphate production from SO 2 continues while CO is inert at these time scales. If SO 2 emissions were the same those two days, then the additional 320 to 370 g s −1 represents about 8 % of the SO 2 in the roughly 5 h since emission, still more than can be attributed to 15 reaction with 0.1 pptv OH.
The correlation betweens OA and SO = 4 at 10 and 180 km and during the P3-B downwind leg reinforce the idea that SO = 4 production governs secondary OA formation through either acid-catalyzed polymerization or organosulfate production. The relative isolation of the oil sands plume and its high acidity may provide an excellent opportunity 20 to study these reactions outside of a laboratory environment.
The increased small particle scattering in Fig. 10b is due to growth of particles initially too small to affect visible light. This is revealed by comparing DMA distributions obtained in each plume. In Fig. 11 we compare number distributions measured near the source (10 km downwind, black) with several distributions measured at 172 to 192 km 25 downwind (red). The mean of the latter (heavy dashed red) has a peak near 0.06 µm compared to 0.015 µm near the source. This indicates that some sizes near the source were too small to be active as CCN in BL clouds but have grown to be effective CCN (i.e. larger than about 0.06 µm) downwind for low clouds. Corresponding growth in the larger mode near 0.1 µm is also evident but as most in this mode were also larger than 0.06 µm before aging, this growth strongly affects light scattering but has only a second order effect on their effectiveness as CCN. As shown in Fig. 3, the DC-8 also intercepted the plume on 29 June. It was inadvertent; the plan was to sample a fire plume seen the previous day. The low-altitude 5 sampling leg was cut short when no smoke was found. However, elevated SO 2 and NO y suggest the oil sands plume was present Fig. 12. A variety of hydrocarbons were also detected, consistent with those identified in the oil sands plume by Simpson et al. (2010). This was not a forest fire plume-CO was only slightly enriched and other biomass indicators like HCN were not elevated at all. The gas ratios did not resemble those identified in the Simpson et al. (2011) analysis of boreal forest fire plumes.
Low-altitude winds and back trajectories are consistent with transport from the Ft. McMurray area (Fig. 13). Trajectories from the southern end of the leg did pass directly over the upgraders about 5 h earlier, but at an altitude 1400 m above the facilities. Low altitude winds evidently carried the plume north for a few hours before lofting it to 900 m, 15 where the back trajectories led to the northern end of the flight leg.
However, the aerosol signature usually present in the oil sands plume were weak. CN and scattering rose only slightly. More surprisingly, while AMS OA correlated well with SO 2 , NO y , and organic vapors, the OA : SO = 4 ratio was 8 : 1 rather than 1 : 1. This suggests some removal mechanism, rather than simply dilution, was active. Rain was 20 reported at Ft. McMurray at 02:00 and 03:00 LT (09:00 and 10:00 UTC) on the 29th (Environment Canada, 2008b), about 7 h before the DC-8 flight. One likely explanation is that much of the SO 2 reacted in cloud to form SO = 4 , which was effectively removed by rain, while leaving insoluble organic vapors unaffected. Photochemical reactions after sunrise could then produce the organic aerosol seen in Fig. 12

Conclusions
While 2 flybys and 3 incidental plume penetrations can only provide short-term estimates of the aerosol output of the Alberta oil sands mining and upgrading operations, these flights have established that emissions fluxes can be measured, at least under favorable weather conditions, but with uncertainties of about a factor of 2. The primary 5 sources of error are mixed layer depth and homogeneity and exchange with overlying air. Careful flight planning, extended sampling including multiple altitudes, and LIDAR data along the flux cross section could greatly enhance the accuracy of the flux calculations, probably to within 50 %. SO 2 and NO 2 emission inventories lie within error estimates of our calculated flux, 10 though our measurements of CO fluxes substantially exceeded reported emissions during two plume penetrations. CO 2 emissions were also above reported emissions, but not by a large factor. Black carbon fluxes are highly uncertain, but appear to be roughly 10 g s −1 .
While neither the mining operations nor the industrial facilities produce much particu-15 late organic matter directly, organic aerosol appears rapidly within the industrial plume in a short time, creating particulate mass approximately equal to the sulphate. Some of this may be primary OA, due to condensation of vapors as the plume cools, but reactions between H 2 SO 4 and either biogenic or plume-derived organic vapors appear to be important. At least over short ranges, we confirm the SO 2 photochemical loss rates 20 from Cheng et al. (1987), though reaction with OH appears insufficient to account for the conversion rate to SO = 4 . The industrial plume has tremendously high aerosol number concentrations, saturating our counting instruments, but those particles are too small to scatter light or serve as effective CCN. They do coagulate and grow, so by the time the plume is a few  B-Lasers O., 67, 399-410, doi:10.1007B-Lasers O., 67, 399-410, doi:10. /s003400050523, 1998 Cheng, L., Peake, E., and Davis, A.: The rate of SO 2 to sulfate particle formation in an air parcel from an oil sands extraction plant plume, JAPCA J. Air Waste Ma., 37, 163-167, 5 doi:10.1080Ma., 37, 163-167, 5 doi:10. /08940630.1987Ma., 37, 163-167, 5 doi:10. .10466211, 1987     10 km away, 6/29 close pass, 7/10 Fig. 9. Relationship between BC mass and aerosol light absorption from the P-3 from directly over the upgrader facilities (10 July) and from about 10 km downwind. 1 s data have been smoothed with a 21 point box filter. and periods on 10 July where the PSAP fluctuated wildly due to humidity transitions in the plumes have been eliminated. The dotted lines are not fits to the data; they are shown to illustrate that the data obtained were similar to literature values for uncoated (8 m 2 g −1 ) and coated (15 m 2 g −1 ) soot. Introduction  Fig. 10. P-3B crossings of the oil sands plume at A) about 10 km (on 6/29) and B) about 180 km (6/28) from the upgraders. Vertical scales are the same on both plots except for CO, BC, and absorption. Absorption is for 530 nm light. The gap in CO data around 22:42 on 6/29 was due to an automatic calibration cycle. Altitude on 6/28 was constant at 1200 m, so is not shown. Low BC and absorption rendered the data sufficiently noisy that a 21 second box filter was applied to make changes visible. The gray region shows the minimal plume period used to calculate fluxes in Table 1, defined by the enhanced SO = 4 . The vertical dashed line marks the end of the maximum plume period, determined from BC and the 1 µm mode visible in the APS data. 36 Fig. 10. P-3B crossings of the oil sands plume at (A) about 10 km (on 29 June) and (B) about 180 km (28 June) from the upgraders. Vertical scales are the same on both plots except for CO, BC, and absorption. Absorption is for 530 nm light. The gap in CO data around 22:42 UTC on 29 June was due to an automatic calibration cycle. Altitude on 28 June was constant at 1200 m, so is not shown. Low BC and absorption rendered the data sufficiently noisy that a 21 s box filter was applied to make changes visible. The gray region shows the minimal plume period used to calculate fluxes in Table 1, defined by the enhanced SO